1: \documentclass[12pt,thmsa]{article}
2: \usepackage{amsmath, latexsym, amsfonts, amssymb, amsthm, amscd}
3: \usepackage{graphics}
4: \textheight 230mm \topmargin 0cm \textwidth 155mm \headheight 0pt
5: \oddsidemargin 0.5cm\headsep 0in
6:
7: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
8:
9: \newtheorem{theorem}{Theorem}
10: \newtheorem{corollary}{Corollary}
11: \newtheorem{proposition}{Proposition}
12: \newtheorem{lemma}{Lemma}
13: \newtheorem{definition}{Definition}
14: \newtheorem{rem}{Remark}
15: \newtheorem{defi}{Definition}
16: \newcommand{\p}{\Bbb{P}}
17: \newcommand{\px}{\Bbb{P}_x}
18: \newcommand{\e}{\Bbb{E}}
19: \newcommand{\ex}{\Bbb{E}_x}
20: \newcommand{\eq}{\Bbb{E}^{\Bbb{Q}}}
21: \newcommand{\eqx}{\Bbb{E}^{\Bbb{Q}}_x}
22: \newcommand{\n}{\underline{n}}
23: \newcommand{\pf}{\Bbb{P}^\uparrow}
24: \newcommand{\pfx}{\Bbb{P}_{x}^\uparrow}
25: \newcommand{\ee}{\mbox{{\bf e}}/\varepsilon}
26: \newcommand{\ix}{\underline{X}}
27: \newcommand{\en}{\mbox{\rm I\hspace{-0.02in}N}}
28: \newcommand{\ind}{\mbox{\rm 1\hspace{-0.04in}I}}
29: \newcommand{\N}{\mbox{\rm I\hspace{-0.02in}N}}
30: \newcommand{\R}{\mbox{\rm I\hspace{-0.02in}R}}
31: \newcommand{\F}{\mbox{\rm I\hspace{-0.02in}F}}
32: \newcommand{\poxi}{\mbox{\raisebox{-1.1ex}{$\rightarrow$}\hspace{-.12in}$\xi$}}
33: \newcommand{\taup}{\mbox{\raisebox{-1.1ex}{$\rightarrow$}\hspace{-.12in}$\tau$}}
34: \newcommand{\ed}{\stackrel{(d)}{=}}
35: \newcommand{\el}{\stackrel{\mathcal L}{=}}
36: \newcommand{\nel}{\stackrel{\mathcal L}{\neq}}
37: \newcommand{\eas}{\stackrel{\mbox{a.s.}}{=}}
38: \newcommand{\eqdef}{\stackrel{\mbox{\tiny$($def$)$}}{=}}
39: \newcommand{\eqas}{\stackrel{\mbox{\tiny$($a.s.$)$}}{=}}
40: \newcommand{\scr}{\scriptstyle}
41: \def\QED{\hfill\vrule height 1.5ex width 1.4ex depth -.1ex \vskip20pt}
42: \renewcommand{\theequation}{\thesection.\arabic{equation}}
43:
44: \begin{document}
45:
46: \title{Reflection principle and Ocone martingales.}
47:
48: \maketitle
49:
50: \begin{center}
51: {\large L. Chaumont}\footnote{$^{,2}$ LAREMA, D\'epartement de
52: Math\'ematiques, Universit\'e d'Angers, 2, Bd Lavoisier - 49045,
53: \\\hspace*{.4in}{\sc Angers Cedex 01.}
54:
55: \hspace*{.05in}$^1$E-mail: loic.chaumont@univ-angers.fr$\;\;\;$
56: $^2$E-mail: lioudmila.vostrikova@univ-angers.fr}{\large
57: and L. Vostrikova$^2$}
58: \end{center}
59: \vspace{0.2in}
60:
61: \begin{abstract} Let $M =(M_t)_{t\geq 0}$ be any continuous real-valued stochastic
62: process. We prove that if there exists a sequence $(a_n)_{n\geq 1}$ of real
63: numbers which converges to 0 and such that $M$ satisfies the
64: reflection property at all levels $a_n$ and $2a_n$ with $n\geq 1$, then $M$ is an
65: Ocone local martingale with respect to its natural filtration.
66: We state the subsequent open question: is this result still true
67: when the property only holds at levels $a_n$~? Then we prove that the later question is
68: equivalent to the fact that for Brownian motion, the $\sigma$-field of the invariant events
69: by all reflections at levels $a_n$, $n\ge1$ is trivial.
70: We establish similar results for skip free $\mathbb{Z}$-valued processes and use them
71: for the proof in continuous time, via a discretisation in space.\\
72:
73: \noindent {\sc Key words and phrases}: Ocone martingale, skip free process, reflection
74: principle, quadratic variation,
75: Dambis-Dubins-Schwarz Brownian motion.\\
76:
77: \noindent MSC 2000 subject classifications: 60G44, 60G42, 60J65.
78: \end{abstract}
79:
80: %\vspace{0.3in}
81: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
82: \section{Introduction and main results}%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
83: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
84:
85: Local martingales whose law is invariant under any integral transformations preserving their
86: quadratic variation were first introduced and characterized by Ocone \cite{oc}.
87: Namely a continuous real-valued local martingale
88: $M=(M_t)_{t\geq 0}$ with natural filtration
89: $\mathbb F = ({\cal F}_t)_{t\geq 0}$ is called {\it Ocone}
90: if
91: \begin{equation}\label{a}
92: \left(\int_0^t H_s dM_s\right)_{t\geq 0}
93: \stackrel{\mathcal L}{=} M\,,
94: \end{equation}
95: for all processes $H$ belonging to the set
96: $$\mathcal H = \{ H=(H_t)_{t\geq 0}\,\,|\,H \mbox{\,is\, $\mathbb F$-predictable},\, |H_t|
97: =1,\, \mbox{\,for \,all\,\,} t\geq 0\}.$$
98:
99: In the primary paper \cite{oc}, the author
100: proved that a local martingale is Ocone whenever it satisfies
101: (\ref{a}) for all processes $H$ belonging to the smaller class
102: of deterministic processes:
103: \begin{equation}
104: \label{b} \mathcal H_1=\{\,\left(\ind _{[0,u]}(t) - \ind
105: _{]u,+\infty [}(t)\right)_{t\geq 0}\,,\,\mbox{with\,\,}u\geq 0\}\,.
106: \end{equation}
107: A natural question for which we sketch out an answer in this paper
108: is to describe minimal sub-classes of $\mathcal H$ characterizing
109: Ocone local martingales through relation (\ref{a}). For instance, it is
110: readily seen that the subset $\{\,\left(\ind _{[0,u]}(t) - \ind
111: _{]u,+\infty [}(t)\right)_{t\geq 0}\,,\,\mbox{with\,\,}u\in E\}$ of
112: $\mathcal H_1$ characterizes Ocone martingales if and only if $E$ is
113: dense in $[0,\infty)$. Let us denote by
114: $\langle M\rangle $ the quadratic variation of $M$. In \cite{oc} it was shown
115: that for continuous local martingales, (\ref{a}) is equivalent to the fact
116: that conditionally to the $\sigma $-algebra $\sigma\{ \langle M
117: \rangle _s, s \geq 0 \}$, $M$ is a gaussian process with independent
118: increments. Hence {\it a continuous Ocone local martingale is a Brownian motion time changed
119: by any independent nondecreasing continuous process.} This is actually the definition
120: we will refer to all along this paper.
121:
122: When the continuous local martingale $M$ is divergent, i.e. $\p$-a.s.
123: \[\lim_{t\rightarrow\infty}\langle M\rangle_t = +\infty\,,\]
124: we denote by $\tau$ the right continuous inverse of $\langle
125: M\rangle$, i.e. for $t\ge0$,
126: $$\tau_t=\inf\{s\ge0:\langle M\rangle_s>t\}\,,$$
127: and we recall that the Dambis-Dubins-Schwarz Brownian motion
128: associated to $M$ is the $({\cal F}_{\tau_t})$-Brownian motion
129: defined by
130: $$B^M\eqdef (M_{\tau_t})_{t\ge0}.$$
131: Then Dubins, Emery and Yor \cite{dey} refined Ocone's characterization
132: by proving that (\ref{a}) is equivalent to each of the following
133: three properties:
134: \begin{itemize}
135: \item[(i)] The processes $\langle M\rangle$ and
136: $B^M$ are independent.
137: \item[(ii)] For every $\mathbb F$-predictable process $H$, measurable
138: for the product $\sigma$-field ${\cal B}(\mathbb{R}_+)\otimes\sigma(\langle
139: M\rangle)$ and such that $\int_0^\infty H^2_s\,d\langle
140: M\rangle_s<\infty$, $\p$-a.s.,
141: \[\e\left(\exp\left(i\int_0^\infty H_s\,dM_s\right)\,|\,\langle
142: M\rangle\right)=\exp\left(-\frac12\int_0^\infty H^2_s\,d\langle
143: M\rangle_s\right)\,.\]
144: \item[(iii)] For every deterministic function $h$ of the form
145: $\sum_{j=1}^n\lambda_j\ind_{[0,a_j]}$,
146: \[\e\left[\exp\left(i\int_0^\infty h(s)\,dM_s\right)\right]=
147: \e\left[\exp\left(-\frac12\int_0^\infty h^2(s)\,d\langle
148: M\rangle_s\right)\right]\,.\]
149: \end{itemize}
150: It can be easily shown that the equivalence between (\ref{a}) and (i), (ii), (iii) also holds
151: in the case when $M$ is not necessarily
152: divergent. This fact will be used in the proof of Theorem \ref{main1}.
153: We also refer to \cite{vy} for further results
154: related to Girsanov theorem and different classes of martingales.
155:
156: In \cite{dey}, the authors conjectured that the class $\mathcal H_1$ can be reduced to
157: a single process, namely that (\ref{a}) is equivalent to:
158: \begin{equation}\label{sgn}
159: \left(\,\int_0^{t}\mbox{sign}(M_s)\,dM_s\,\right)_{\,t\ge0}\el M\,.
160: \end{equation}
161: In fact, (\ref{sgn}) holds if and only if $B^M$ and
162: $\langle M \rangle $ are conditionally independent given
163: the $\sigma$-field of
164: invariant sets by the L\'evy transform of $B^M$, i.e.
165: $B^M\mapsto \left(\int_0^{\cdot}\mbox{sign}(B^M_s)\,dB^M_s\right)$, see \cite{dey}.
166: Hence if the L\'evy transform of Brownian motion is ergodic, then
167: $B^M$ and $\langle M \rangle$ are independent and (\ref{sgn})
168: implies that $M$ is an Ocone local martingale. The converse is also proved in \cite{dey}, that is
169: if (\ref{sgn}) implies that $M$ is an Ocone local martingale, then the L\'evy transform of
170: Brownian motion is ergodic.
171:
172: Different other approaches have been proposed to prove
173: ergodicity of the L\'evy transform but this problem is still open.
174: Among the most accomplished works in this direction, we may cite
175: papers by Malric \cite{ma1}, \cite{ma2} who studied the density of
176: zeros of iterated L\'evy transform. Let us also mention
177: that in discrete time case this problem has been
178: treated in \cite{ds} where the authors proved that an equivalent of the L\'evy transform for
179: symmetric Bernoulli random walk is ergodic.\\
180:
181: In this paper we exhibit a new sub-class of $\mathcal H_1$
182: characterizing continuous Ocone local martingales which is related to
183: first passage times and the reflection property of stochastic
184: processes. If $M$ is the standard Brownian motion and $T_a(M)$ the
185: first passage time at level $a$, i.e.
186: \begin{equation}\label{ta}
187: T_a(M)=\inf\{t\geq 0:M_t=a\},
188: \end{equation}
189: where here and in all the remainder of this article, we make the convention that
190: $\inf \{\emptyset\}= +\infty$, then for all $a\in\mathbb{R}$:
191: $$(M_t)_{t\geq 0}\el ( M_t\ind _{\{ t\leq T_a(M)\}} +
192: (2a-M_t)\ind _{\{t > T_a(M)\}})_{t\geq 0}.$$
193: It is readily checked that this identity in law actually holds for any
194: continuous Ocone local martingale. This property is
195: known as the {\it reflection principle at level $a$} and was first observed
196: for symmetric Bernoulli random walks by Andr\'e \cite{an}. We will use
197: this terminology for any continuous stochastic process $M$ and when
198: no confusion is possible, we will denote by $T_a=T_a(M)$ the first passage
199: time at level $a$ by $M$ defined as above.
200:
201: Let $(\Omega ,\mathcal F , \mathbb F, \p)$ be the canonical space of
202: continuous functions endowed with its natural right-continuous filtration $\mathbb F =
203: (\mathcal F _t)_{t\geq 0}$ completed by negligible sets of $\mathcal
204: F= \bigvee_{t\geq 0} \mathcal F_t$.
205: The family of transformations $\Theta^a$, $a\ge0$, is
206: defined for all continuous functions $\omega\in\Omega$ by
207: \begin{equation}\label{transf}
208: \Theta^{a}(\omega)=(\omega_t\ind _{\{ t\leq T_a\}} +
209: (2a-\omega_t)\ind _{\{t > T_a\}})_{t\geq 0}\,.
210: \end{equation}
211: Note that
212: $\Theta^a(\omega)=\omega$ on the set $\{\omega : T_a(\omega)=\infty\}$. When
213: $M$ is a local martingale, $\Theta^{a}(M)$ can by expressed in terms
214: of a stochastic integral, i.e.
215: \[\Theta^{a}(M)=\left(\int_0^t\,\left(\ind _{[0,T_a]}(s) - \ind
216: _{]T_a,+\infty [}(s)\,\right)\,dM_s\right)_{t\ge0}.\]
217: The set
218: $\mathcal H_2 = \{\left(\ind _{[0,T_a]}(t) -
219: \ind_{]T_a,+\infty[}(t)\right)_{t\ge0}\,|\,a\ge0\}$ is a subclass of
220: $\mathcal H$ which provides a family of transformations preserving
221: the quadratic variation of $M$ and we will prove that it characterizes Ocone local
222: martingales. But the fact that the transformations $\omega\mapsto \Theta^{a}(\omega)$ are defined
223: for all continuous functions $\omega\in\Omega$ allows us to
224: characterize Ocone local martingales in the whole set of continuous stochastic
225: processes as shows our main result.
226:
227: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
228: \begin{theorem}\label{main1}
229: Let $M =(M_t)_{t\geq 0}$ be a continuous stochastic process defined on the canonical probability
230: space, such that $M_0=0$.
231: If there exists a sequence $(a_n)_{n\geq 1}$ of positive real numbers
232: such that $\lim_{n\rightarrow \infty} a_n=0$ and for all
233: $n\ge0$:
234: \begin{equation}\label{hyp}
235: \Theta^{a_n}(M)\el \Theta^{2a_n}(M)\el M\,,
236: \end{equation}
237: then $M$ is an Ocone local martingale with
238: respect to its natural filtration. Moreover, if $T_{a_1}<\infty$
239: a.s., then $M$ is a divergent local martingale.
240: \end{theorem}
241:
242: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
243: \begin{rem} It is natural to wonder about the necessity of the
244: hypothesis $\Theta^{2a_{n}}(M)\el M$ in Theorem $\ref{main1}$. The discrete time counterpart
245: of this problem which is presented in section $\ref{skipfree}$, shows that it is necessary
246: for a skip free process $M$ to satisfy $\Theta^a(M)\el M$, for $a=0,1$
247: and $2$ in order to be a skip free Ocone local martingale, i.e. the reflection property
248: at $a=0$ and $1$ is not sufficient, see the counterexamples in section $\ref{counterexample}$.
249: This argument seems to confirm that the assumption $\Theta^{2a_{n}}(M)\el M$ is necessary
250: in continuous time.
251: \end{rem}
252:
253: In an attempt to identify the sequences $(a_n)_{n\geq 1}$ which characterize Ocone local martingales,
254: we obtained the following theorem. Let
255: $a=(a_n)_{n\geq 1}$ be a sequence of real numbers with
256: $\lim_{n\rightarrow\infty} a_n=0$ and
257: let ${\mathcal I}^a$ the sub-$\sigma$-field of
258: the invariant sets by all the transformations $\Theta^{a_n}$, i.e.
259: \[{\cal I}^{a}=\{F\in{\cal F}:\ind_{F}\circ\Theta^{a_n}\stackrel{\mbox{\small a.s.}}{=}
260: \ind_F,\;\;\mbox{for all $n\ge0$}\}.\]
261:
262: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
263: \begin{theorem}\label{main2}
264: The following assertions are equivalent:
265: \begin{itemize}
266: \item[$(i)$] Any continuous local martingale $M$ satisfying
267: $\Theta^{a_n}(M)\el M$ for all $n\ge0$ is an Ocone local martingale.
268: \item[$(ii)$] The sub $\sigma$-field
269: ${\cal I}^{a}$ is trivial for the Wiener measure on the canonical space $(\Omega ,\mathcal F)$.
270: \end{itemize}
271: \end{theorem}
272: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
273:
274: \begin{rem} It follows from Theorems $\ref{main1}$ and $\ref{main2}$ that if the sequence
275: $(a_n)$ contains a subsequence $(2a_{n'})$ $( $this holds, for instance, when $(a_n)$
276: is dyadic sequence $)$, then the sub $\sigma$-field
277: ${\cal I}^{a}$ is trivial for the Wiener measure on $(\Omega,\mathcal F)$. So, our open question
278: is equivalent to: is the sub $\sigma$-field ${\cal I}^{a}$ trivial for any sequence $(a_n)$
279: decreasing to zero~?
280: \end{rem}
281:
282: In the next section, we prove analogous results for skip
283: free processes. We use them as preliminary results to prove Theorem
284: \ref{main1} in section \ref{proof}. In section \ref{counterexample}, we give
285: counterexamples in the discrete time setting, related to Theorem \ref{prop2}.
286: Finally, in section \ref{conj}, we prove Theorem \ref{main2}.
287:
288:
289: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
290: \section{Reflecting property and skip free processes}\label{skipfree}
291: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
292: \subsection{Discrete time skip free processes}
293: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
294: A discrete time skip free process $M$ is any measurable
295: stochastic process with $M_0=0$ and for all $n\ge1$,
296: $\Delta M_n= M_{n}-M_{n-1}\in\{-1,0,1\}$. This section is devoted to an analogue of
297: Theorem \ref{main1} for skip free processes.\\
298:
299: To each skip free process $M$, we associate the increasing process
300: $$[M]_n = \sum_{k=0}^{n-1}(M_{k+1}-M_k)^2\,,\;\;n\ge1\,,\;\;\;[M]_0=0\,,$$
301: which is called the {\it quadratic variation} of $M$. In this section,
302: since no confusion is possible, we will use the same notations for discrete
303: processes as in continuous time case. For every integer $a\ge0$, we denote
304: by $T_a$ the first passage time by $M$ to the level $a$,
305: $$T_a=\inf\{k\geq 0:M_k=a\}\,.$$
306: We also introduce the inverse process $\tau$ which is defined by $\tau_0=0$ and for $n\geq 1$,
307: $$\tau_n=\inf\{k>\tau_{n-1}:[M]_k=n\}$$
308: with $\inf\{\emptyset\} = \tau_{n-1}$.
309: Then we may
310: define
311: \begin{equation}\label{sup}
312: S^M=(M_{\tau_n})_{n\ge0}
313: \end{equation}
314: Denote also
315: $$T=\inf\{k\geq 0 \,:\, [S^ M]_k=[S^ M]_{\infty}\}\,,$$
316: then note that the paths of $S^M$ are such that $\Delta S^M_k\in\{-1,+1\}$, for all $k\leq T$
317: and $\Delta S_k^M=0$, for all $k>T$, where $\Delta S_k^M= S_k - S_{k-1}^M$.\\
318:
319: We recall that skip free martingales are just skip free processes being martingales
320: with respect to some filtration.
321: It is well known that for any divergent free skip martingale $M$, that is satisfying
322: $\lim_{n\rightarrow+\infty}[M]_n=+\infty$, a.s., the process $S^M$ is a symmetric
323: Bernoulli random walk on $\mathbb Z$. This property is the equivalent of the
324: Dambis-Dubins-Schwartz theorem for continuous martingales. In discrete
325: time, the proof is quite straightforward and we recall it now.
326:
327: A first step is the equivalent of L\'evy's characterization for skip free martingales~:
328: any skip free martingale $S$ such that $S_{n+1}-S_n\neq0$, for all
329: $n\ge0$ (or equivalently, whose quadratic variation satisfies
330: $[S]_n=n$) is a symmetric Bernoulli random walk. Indeed for $n\ge1$,
331: $S_{1},S_{2}-S_{1},\dots,S_{n}-S_{n-1}$ are i.i.d. symmetric Bernoulli
332: r.v.'s
333: if and only if for any subsequence $1\le n_1\le\dots\le n_k\le n$:
334: \begin{eqnarray*}
335: &&\e[(S_{n_1}-S_{n_1-1})(S_{n_2}-S_{n_2-1})\dots(S_{n_k}-S_{n_k-1})]=\\
336: &&\quad\qquad\e[S_{n_1}-S_{n_1-1}]\e[S_{n_2}-S_{n_2-1}]\dots\e[S_{n_k}-S_{n_k-1}]=0
337: \end{eqnarray*}
338: and this identity can be easily checked from the martingale property.
339: Finally call ${\mathbb F}=({\cal F}_n)_{n\ge0}$ the natural filtration generated by $M$.
340: Since $[M]_n$ is an ${\mathbb F}$-adapted process, from the optional stopping theorem,
341: $S^M$ is a martingale with respect to the filtration $({\cal
342: F}_{\tau_n})_{n\ge0}$ and since its increments cannot be $0$, we
343: conclude from L\'evy's characterization.
344:
345: We recall also the following important property:
346: any skip free process which is a symmetric Bernoulli random walk time
347: changed by an independent nondecreasing skip free process,
348: is a local martingale with respect to its natural filtration.\\
349:
350: This leads to the definition:
351: \begin{definition}\label{def}
352: A discrete Ocone local martingale is a symmetric Bernoulli random walk time changed by any independent
353: increasing skip free process.
354: \end{definition}
355:
356: We emphasize that in this particular case, Definition \ref{def}
357: coincides with the general definition of Ocone \cite{oc}.
358: It should also be noticed that the symmetric Bernoulli random walk of Definition 1
359: is not necessarily the same as in (\ref{sup}). It coincides with $S^M$ if $M$ is a divergent process.
360: If $M$ is not divergent, then it can obtained obtained from the initial one by pasting of
361: an independent symmetric Bernoulli random walk (see Lemma 3), otherwise the independence can fail.
362: \vspace{0.2cm}
363:
364: \noindent A counterpart of transformations $\Theta^a$ defined in (\ref{transf}) for skip free processes
365: is given for all integers $a\ge0$ by
366: \begin{equation}\label{trdisc}
367: \Theta^a(M)_n=\sum_{k=1}^n(\ind_{\{k \leq T_a\}}- \ind_{\{k >
368: T_a\}})\Delta M_{k}\,,
369: \end{equation}
370: where $\Delta M_k=M_k-M_{k-1}$. Again in the following discrete time counterpart
371: of Theorem~\ref{main1}, we characterize discrete Ocone local martingales in the whole set of skip free
372: processes.
373:
374: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
375: \begin{theorem}\label{prop2}
376: Let $M$ be any discrete skip free process. Assume that for all $a\in\{0,1,2\}$,
377: \begin{equation}\label{identity}
378: \Theta^a(M)\el M,\end{equation}
379: then $M$ is a discrete Ocone local martingale with respect to its
380: natural filtration. If in addition $T_1 < \infty $ a.s. then $M$ is a
381: divergent local martingale.
382: \end{theorem}
383: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
384:
385: \noindent The proof of Theorem \ref{prop2} is based on the following
386: crucial combinatorial lemma concerning the set of sequences of
387: partial sums of elements in $\{-1,+1\}$ with length $m\ge1$:
388: \[\Lambda^m=\{(s_0,s_1,\dots,s_m):s_0=0\;\;\mbox{and}\;\; \Delta s_{k}
389: \in\{-1,+1\} \,\mbox{for}\,\;\; 1\leq k\leq m\,\},\]
390: where $\Delta s_k=s_k - s_{k-1}$.
391:
392: For each sequence $s\in\Lambda^m$, and each integer $a$, we define
393: $T_a(s)=\inf\{k\geq 0:s_k=a\}$, with $\inf\emptyset=+\infty$. The transformation $\Theta^a(s)$ is defined
394: for each $s\in\Lambda^m$ by
395: \[\Theta^a(s)_n=\sum_{k=1}^n(\ind_{\{k\leq T_a(s)\}}- \ind_{\{k >
396: T_a(s)\}})\Delta s_{k}\,,\;\;\;n\le m.\]
397:
398: %%%%%%%%%%%%%%%%%%%%
399: \begin{lemma}\label{lem1}Let $m\geq1$ be fixed.
400: For any two elements $s$ and $s'$ of the set $\Lambda^m$ such that
401: $s\neq s'$, there are integers $a_1,a_2,\dots,a_k\in\{0,1,2\}$
402: depending on $s$ and $s'$ such that
403: \begin{equation}\label{ssp}
404: s'=\Theta^{a_k}\Theta^{a_{k-1}}\dots\Theta^{a_1}(s)\,.
405: \end{equation}
406: Moreover, the integers $a_1,\dots,a_k$ can be chosen so that
407: $s\in\Lambda_{a_1}^m$ and $\Theta^{a_{i-1}}\Theta^{a_{i-2}}\dots\Theta^{a_{1}}(s)\in\Lambda_{a_i}^m$,
408: for all $i=2,\dots,k$ where
409: \[\Lambda_a^m=\{s\in\Lambda^m,\,T_a(s)\le m-1\}\,.\]
410: \end{lemma}
411: %%%%%%%%%%%%%%%%%%%
412:
413:
414: \noindent {\it Proof}. The last property follows from the simple remark
415: that for $s\in\Lambda^m$ we have that $\Theta^a(s)\neq s$ if and only if
416: $s\in\Lambda_a^m$. So, for the rest of the proof we suppose that all
417: transformations used verify the above property.
418:
419: Let $\bar{s}^{(m)}$ be the sequence of
420: $\Lambda_m$ defined by $\bar{s}_1^{(m)}=1$ and
421: $\Delta\bar{s}_{k}^{(m)}= - \Delta\bar{s}_{k-1}^{(m)}$ for all $2\le k\le m$.
422: That is $\bar{s}^{(m)}\eqdef (0,1,0,1,\dots,0,1)$ if $m$ is odd and
423: $\bar{s}^{(m)}\eqdef (0,1,0,1,\dots,1,0)$ if $m$ is even.
424:
425: First we prove that
426: the statement of the lemma is equivalent to the following one: for
427: any sequence $s$ of $\Lambda_m$ such that $s\neq \bar{s}^{(m)}$, there
428: are integers $b_1,b_2,\dots,b_p\in\{0,1,2\}$ such that
429: \begin{equation}\label{sbar}
430: \bar{s}^{(m)}=\Theta^{b_p}\Theta^{b_{p-1}}\dots\Theta^{b_1}(s)\,.
431: \end{equation}
432: Indeed, suppose that the later property holds and let
433: $s'\in\Lambda_m$ such that $s'\neq s$. If $s'=\bar{s}^{(m)}$, then the
434: sequence $b_1,b_2,\dots,b_p$ satisfies the statement of the lemma.
435: If $s'\neq \bar{s}^{(m)}$, then let $c_1,\dots,c_l\in\{0,1,2\}$ such
436: that
437: \[\bar{s}^{(m)}=\Theta^{c_l}\Theta^{c_{l-1}}\dots\Theta^{c_1}(s')\,.\]
438: We notice that the transformations $\Theta^a$ are involutive, i.e. for all
439: $x\in\Lambda_m$,
440: \begin{equation}
441: \label{r} \Theta^a\Theta^a(x)=x.
442: \end{equation}
443: Then we have
444: $\Theta^{c_1}\Theta^{c_{2}}\dots\Theta^{c_l}(\bar{s}^m)=s'$, so that
445: \[s'=\Theta^{c_1}\Theta^{c_{2}}\dots\Theta^{c_l}
446: \Theta^{b_{p}}\Theta^{b_{p-1}}\dots\Theta^{b_1}(s)\,,\] which
447: implies (\ref{ssp}). The fact that (\ref{ssp}) implies (\ref{sbar}) is
448: obvious.\\
449:
450: Now we prove (\ref{sbar}) by induction in $m$.
451: It is not difficult to see that the result is
452: true for $m=1,2$ and $3$. Suppose that the result is true up to $m$ and
453: let $s\in\Lambda^{m+1}$ such that $s\neq\bar{s}^{(m+1)}$. For $j\le
454: m$, we call $s^{(j)}$ the truncated sequence
455: $s^{(j)}=(s_0,s_1,\dots,s_j)\in\Lambda^j$.
456: From the hypothesis of induction, there exist
457: $b_1,b_2,\dots,b_p\in\{0,1,2\}$ such that
458: \begin{equation}\label{eqlem1}
459: \bar{s}^{(m)}=\Theta^{b_p}\Theta^{b_{p-1}}\dots\Theta^{b_1}(s^{(m)})\,
460: \end{equation}
461: where
462: \begin{equation}\label{eqlem2}
463: \mbox{$s^{(m)}\in\Lambda_{b_1}^m\;$ and $\;\Theta^{b_{i-1}}\Theta^{b_{i-2}}\dots
464: \Theta^{b_{1}}(s^{(m)})\in\Lambda_{b_i}^m$,$\;$ for all $i=2,\dots,p$.}
465: \end{equation}
466: Then, let us consider separately the case where $m$ is even and the case where $m$ is odd.
467:
468: If $m$ is even and $\Delta s_m\Delta s_{m+1}=-1$, then we obtain directly that
469: $$\Theta^{b_p}\Theta^{b_{p-1}}\dots\Theta^{b_1}(s)=\bar{s}^{(m+1)}\,.$$
470: Indeed, from (\ref{eqlem2}), none of the transformations $\Theta^{b_{i-1}}\dots\Theta^{b_1}$,
471: $i=2,\dots,p$ affects the last step of $s$, so the identity follows from (\ref{eqlem1}).
472:
473: If $m$ is even and $\Delta s_m\Delta s_{m+1}=1$, then from the hypothesis of induction
474: there exist $d_1,d_2,\dots,d_{r}\in\{0,1,2\}$ such that
475: \begin{equation}\label{eqlem3}\Theta^{d_r}\dots\Theta^{d_1}(s^{(m)})=
476: (\bar{s}^{(m-1)}, 2)
477: \end{equation}
478: which, from the above remark, may be chosen so that
479: \begin{equation}\label{eqlem4}
480: \mbox{$s^{(m)}\in\Lambda_{d_1}^m\;$ and $\;\Theta^{d_{i-1}}\Theta^{d_{i-2}}\dots
481: \Theta^{d_{1}}(s^{(m)})\in\Lambda_{d_i}^m$,$\;$ for all $i=2,\dots,r$.}
482: \end{equation}
483: Since from (\ref{eqlem4}), none of the transformations $\Theta^{d_{i}}\dots\Theta^{d_1}$, $i=1,\dots,r$
484: affects the last step of $s$, it follows from (\ref{eqlem3}) that
485: \begin{equation}\label{eqlem5}\Theta^{d_r}\dots\Theta^{d_1}(s)=
486: (\bar{s}^{(m-1)}, 2, 3)\,.\end{equation}
487: Then by applying transformation $\Theta^2$, we obtain:
488: \begin{equation}\label{eqlem6}\Theta^2(\bar{s}^{(m-1)}, 2, 3)=(\bar{s}^{(m-1)}, 2, 1)\,.
489: \end{equation}
490: Hence, from (\ref{eqlem3}) and since none of the transformations
491: $\Theta^{d_{r-i}}\dots\Theta^{d_r}$, $i=0,1,\dots,r-1$
492: affects the last step of $(\bar{s}^{(m-1)},2,1)$, we have
493: \[\Theta^{d_1}\Theta^{d_2}\dots\Theta^{d_r}(\bar{s}^{(m-1)},2,1)=(s^{(m)},s_m-\Delta s_{m+1})\,.\]
494: Finally from (\ref{eqlem1}) and (\ref{eqlem2}), we have
495: \[\Theta^{b_p}\Theta^{b_{p-1}}\dots\Theta^{b_1}\Theta^{d_1}\dots\Theta^{d_r}\Theta^2\Theta^{d_r}\dots\Theta^{d_1}(s)=\overline{s}^{(m+1)}\]
496: and the induction hypothesis is true at the order $m+1$, when $m$ is even.\\
497:
498: The proof when $m$ is odd is very similar and we will pass over some of the arguments
499: in this case. If $m$ is odd and $\Delta s_m\Delta s_{m+1}=-1$, then we obtain directly that
500: $$\Theta^{b_p}\Theta^{b_{p-1}}\dots\Theta^{b_1}(s)=\bar{s}^{(m+1)}\,.$$
501: If $m$ is odd and $\Delta s_m\Delta s_{m+1}=1$ then from the hypothesis of induction,
502: there exist $d_1,d_2,\dots,d_{r}\in\{0,1,2\}$ such that
503: \begin{equation}\label{eqlem8}
504: \Theta^{d_r}\dots\Theta^{d_1}(s^{(m)})=
505: (\bar{s}^{(m-1)}, -1)\end{equation}
506: and
507: \begin{equation}\label{eqlem9}
508: \mbox{$s^{(m)}\in\Lambda_{d_1}^m\;$ and $\;\Theta^{d_{i-1}}\Theta^{d_{i-2}}\dots
509: \Theta^{d_{1}}(s^{(m)})\in\Lambda_{d_i}^m$,$\;$ for all $i=2,\dots,r$.}
510: \end{equation}
511: Then it follows from (\ref{eqlem8}) and (\ref{eqlem9}) that
512: \begin{equation}\label{eqlem10}
513: \Theta^{d_r}\dots\Theta^{d_1}(s)=
514: (\bar{s}^{(m-1)}, -1, -2)\end{equation}
515: and by performing the transformation $\Theta^1\Theta^0\Theta^1=\Theta^{-1}$,
516: \begin{equation}\label{eqlem11}\Theta^0\Theta^1\Theta^0(\bar{s}^{(m-1)}, -1, -2)=(\bar{s}^{(m-1)}, -1, 0)\,.
517: \end{equation}
518: From (\ref{eqlem8}) and (\ref{eqlem9}), it follows that
519: $$\Theta^{d_1}\dots\Theta^{d_r}(\bar{s}^{(m-1)}, -1, 0)=(s^{(m)},s_m-\Delta s_{m+1}),$$
520: which finally gives from (\ref{eqlem1}) and (\ref{eqlem2}),
521: $$\Theta^{b_p}\Theta^{b_{p-1}}\dots\Theta^{b_1}\Theta^{d_1}\dots\Theta^{d_r}\Theta^0\Theta^1\Theta^0
522: \Theta^{d_r}\dots\Theta^{d_1}(s)=\bar{s}^{(m+1)}$$
523: and ends the proof of the lemma.~\QED
524:
525: In the proof of Theorem \ref{prop2} for technical reasons we have to
526: consider two cases: $T_1<\infty$ a.s. and $\p(T_1 = \infty )>0$.
527: Lemma \ref{lem2} proves that in the first case $M$ is a divergent process.
528:
529: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
530: \begin{lemma}\label{lem2} Any skip free process such that $T_1<\infty$ a.s.~and
531: $\Theta^a(M)\el M$ for $a=0$ and $1$ satisfies$:$
532: $$\lim_{n\rightarrow+\infty}\,[M]_n=+\infty\,,\;\;\;\mbox{a.s.}$$
533: \end{lemma}
534: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
535:
536: \noindent {\it Proof}. Let us introduce the first exit time from
537: the interval $[-a,a]$:
538: \[\sigma_a(M)=\inf\{n:|M_n|=a\}\,,\]
539: where $a$ is any integer. Let us put
540: \[\Psi^a(M)=\left(\sum_{k=1}^n(\ind_{\{k\le\sigma_a\}}- \ind_{\{k>\sigma_a\}})\Delta M_k\right)_{n\geq 0}\,,\]
541: where $\Delta M_k=M_k-M_{k-1}$.
542: First we observe that if $\Theta^a(M)\el M$ for $a=0$ and 1, then $\Psi^a(M)\el M$, for $a=0$ and 1.
543: This assertion is obvious for $a=0$ since $\sigma_0=T_0$. For $a=1$, it follows
544: from the almost sure identity:
545: \[\Psi^{a}(M)=\Theta^a(M)\ind_{\{T_a<T_{-a}\}}+\Theta^{-a}(M)\ind_{\{T_{-a}<T_a\}}\,.\]
546: and the equalities:
547: \begin{eqnarray*}
548: &&\{T_a(M)<T_{-a}(M)\}=\{T_a(\Theta^a(M))<T_{-a}(\Theta^a(M))\},\\\\
549: &&\{T_{-a}(M)<T_{a}(M)\}=\{T_{-a}(\Theta^{-a}(M))<T_{a}(\Theta^{-a}(M))\}\,.\end{eqnarray*}
550: Then from the almost sure inequality
551: \[\sigma_3(\Psi^1(M))\le \max\{T_1(M),T_{-1}(M)\}\,,\]
552: the fact that $T_1(M)<\infty$, $T_{-1}(M)<\infty$ a.s. and the
553: identity in law $\Psi^1(M)\el M$, we deduce that $\sigma_3(M)<+\infty$, a.s.
554: It means, since $M$ is a symmetric
555: process, that $T_3(M)<\infty$ and $T_{-3}(M)<\infty$, a.s. Generalizing the above inequality, we obtain
556: \[\sigma_{a+2}(\Psi^1(M))\le \max\{T_a(M),T_{-a}(M)\}\,.\] This
557: gives in the same manner as before, that for each $a\geq 0$, $\sigma_a < \infty $ a.s.. From this it is
558: not difficult to see that $\lim _{n\rightarrow \infty}[M]_n = +\infty$, $\p$-a.s..
559: \QED
560:
561: The next lemma shows that in the case $\p(T_1=\infty)>0$ we can modify our
562: process $M$ by pasting to it an independent symmetric Bernoulli random
563: walk $S$ and reduce the case $\p(T_1=\infty)>0$ to the case $T_1 <
564: \infty$ a.s..
565:
566: We denote by $[M]_{\infty} = \lim _{k\rightarrow \infty}[M]_k$ which always
567: exists since it is an increasing process and we put
568: $$T=\inf\{k\geq 0:[M]_k=[M]_\infty\}\,,$$
569: with $\inf \{\emptyset \} = +\infty$. We denote the extension of the
570: process $M$ by $X$ where for all $k\geq 0$
571: $$X_k = M_k \ind_{\{ k<T\}} + (M_T + S_{k-T})\ind_{\{k\geq T\}}.$$
572: Note that $X=M$, on the set $\{T=\infty\}$.
573: \begin{lemma}\label{lem3}
574: Let $M$ be a discrete skip free process which satisfies
575: $\Theta^a(M)\el M$ for some $a\in\mathbb{Z}$. Then $X$
576: also satisfies $\Theta^a(X)\el X$. Moreover, the $\sigma-$algebras generated by the respective
577: quadratic variations coincide, i.e.
578: $\sigma([M])=\sigma([X])$, $X$ is a divergent process $\p$-a.s. and $M=S^X_{[M]}$.
579: \end{lemma}\vspace{0.2cm}
580:
581: \noindent {\it Proof}. We show that reflection property holds for
582: $X$. In this aim, we consider the
583: two processes $Y$ and $Z$ such that for all $k\geq 0$,
584: $$Y_k = \Theta^a(M)_k \ind_{\{ k<T\}} + (\Theta^a(M)_T - S_{k-T})\ind_{\{k\geq T\}},$$
585: $$Z_k = M_k \ind_{\{ k<T\}} + (M_T + \Theta^{a-M_T}(S)_{k-T})\ind_{\{k\geq T\}}.$$
586: We remark that
587: \begin{equation}\label{ast}
588: \Theta^a(X)=Y\ind_{\{T_a(Y)\leq T\}}+Z\ind_{\{T_a(Z)> T\}}
589: \end{equation}
590: and we write the same kind of decomposition for $X$:
591: \begin{equation}\label{ast2}
592: X=X\ind_{\{T_a(X)\leq T\}}+X\ind_{\{T_a(X)> T\}}\,.
593: \end{equation}
594: In view of (\ref{ast}) and (\ref{ast2}), to obtain $X\el\Theta^a(X)$
595: it is sufficient to show that for all bounded and measurable functional $F$,
596: $$\e [F(X)]=\e[F(Y)\ind_{\{T_a(Y)\leq T\}}]+\e[F(Z)\ind_{\{T_a(Z)> T\}}]\,.$$
597: Since reflection is a transformation which preserves the quadratic
598: variation of the process, the random time $T$ can be defined as a
599: functional of $Y$ as well as a functional of $Z$.
600: So we see that
601: the last equality is equivalent to $X\stackrel{\mathcal L}{=}Y$ and
602: $X\stackrel{\mathcal L}{=}Z$.
603: The first equality in law follows from the fact that
604: $$ (M,S) \stackrel{\mathcal L}{=}(\Theta^a(M), -S)$$
605: which holds due to the reflection property of $M$ and $S$, and independency of
606: $M$ and $S$. The second one holds since it can be reduced to the reflection property of $S$ itself,
607: by conditioning with respect to $M$.
608:
609: Finally, the identity $M=S^X_{[M]}$ just follows from the construction of $X$.\QED
610:
611:
612: \noindent {\it Proof of Theorem $\ref{prop2}$}. Since both processes $M$ and $\Theta^a(M)$ have the same
613: quadratic variation, the identity in law of the statement is equivalent to:
614: for all $a=0,1,2$
615: \begin{equation*}
616: (M,[M])\el(\Theta^a(M),[M])\,.
617: \end{equation*}
618: Then we remark that the above equalities are equivalent to: for all $a=0,1,2$
619: \begin{equation*}
620: (S^M,[M])\el(S^{\Theta^a(M)},[M])\,.
621: \end{equation*}
622: Now it is crucial to observe the path by path equality: for each $a=0,1,2$
623: \[S^{\Theta^a(M)}=\Theta^a(S^M)\,,\]
624: from which we obtain
625: \begin{equation}\label{eq1p}
626: (S^M,[M])\el(\Theta^a(S^M),[M])\,.
627: \end{equation}
628: Hence,
629: \begin{equation}
630: \label{eq3p} \mathcal L( S^M | [M]) = \mathcal L(\Theta ^a(S^M) |
631: [M])\end{equation}
632: Fix $m\ge1$ and let $s,s'\in\Lambda^m$ with $s\neq s'$ be fixed. Consider the sequence of integers
633: $a_1,a_2,\dots,a_k\in\{0,1,2\}$ given in Lemma~\ref{lem1} such that
634: \begin{equation}\label{eq2p}
635: s=\Theta^{a_k}\Theta^{a_{k-1}}\dots\Theta^{a_1}(s')\,.
636: \end{equation}
637: Denote by $S^{M,m}$ the restricted path $(S_0,S_1,\dots,S_m)$.
638: Iterating (\ref{eq3p}), we may write for all $u\in \Lambda^m$ :
639: \[\p\left(S^{M,m}=u\,|\,[M]\right)=\p\left(\Theta^{a_1}\Theta^{a_2}\dots\Theta^{a_{k}}
640: (S^{M,m})=u\,|\,[M]\right)\,.\]
641: Applying (\ref{r}), we see that the right-hand side is equal to
642: \[\p\left(S^{M,m}=
643: \Theta^{a_k}\Theta^{a_{k-1}}\dots\Theta^{a_1}(u)\,|\,[M]\right)\,.\]
644: Take now $u=s'$ and use (\ref{eq2p}), to obtain
645: \begin{equation}\label{00}
646: \p\left(S^{M,m}=s'\,|\,[M]\right)=\p\left(S^{M,m}=s\,|\,[M]\right)\,.
647: \end{equation}
648: If $T_1<\infty$ a.s. then from Lemma \ref{lem2} we can see that $M$ is
649: divergent and for all $m\geq 0$ $\p(S^{M,m}\in\Lambda^m)=1$.
650: Then from (\ref{00}) the law of $S^{M,m}$ is uniform
651: over $\Lambda^m$ and it coincides with the conditional law of $S^{M,m}$ given
652: $[M]$. Hence, $S^{M,m}$ is symmetric Bernoulli
653: random walk on $[0, m]$ independent from $[M]$. Since this holds
654: for all $m\ge0$, we conclude that $S^{M}$ is
655: a symmetric Bernoulli random walk which is independent of $[M]$. So from Definition \ref{def},
656: $M$ is a divergent Ocone local martingale.
657:
658: If $\p(T_1 = \infty)>0$, we consider the extension $X$ of the process
659: $M$ defined in Lemma \ref{lem3}. Then, $X$ satisfies the hypotheses of Theorem \ref{prop2}. Moreover,
660: from Lemma \ref{lem3}, $\p(T_1(X)<\infty)=1$. From what has just been proved
661: $S^X$ is a symmetric Bernoulli random walk which is independent of
662: $[X]$, and hence from $[M]$. This implies that $S^X$ and $[M]$ are
663: independent. From Lemma \ref{lem3} we have $M=S^X_{[M]}$, and, hence, the process $M$
664: is itself an Ocone martingale by Definition \ref{def}.\QED
665:
666:
667: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
668: \subsection{Counterexamples}\label{counterexample}
669: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
670:
671: In this part, we give two examples of a discrete skip free process $M$ which satisfy
672: $M_0=0$, $\Theta^0(M)\el M$ and $\Theta^1(M)\el M$, but which are not a discrete
673: Ocone martingales.\vskip 0.5cm
674:
675: \noindent {\bf Counterexample 1}: Let $(\epsilon_k)_{k\geq 1}$ be a sequence of
676: independent symmetric Bernoulli random variables. We put
677: $M_0=0,\, \Delta M_1= \epsilon_1,\, \Delta M_2= \epsilon_2,\, \Delta
678: M_3= \epsilon_2$ and for $k>3$, $ \Delta M_k= \epsilon_k$. We
679: introduce also
680: $$M_n= \sum_{k=1}^n \Delta M_k.$$
681: Since $[M]_n=n$ for all $n\geq 1$ and since $M$ is not Bernoulli
682: random walk, it can not be an Ocone martingale.
683:
684: Let us verify that $\Theta^a(M)\el M$ for $a\in \mathbb N \setminus
685: \{2\}$.
686: For $a=0$ we have reflection property since the $\epsilon _k$'s are
687: symmetric and independent.
688: For $a=1$ we consider four possible cases
689: related with the values of $(M_1, M_2, M_3)$. Let us put $R_n
690: =\sum_{k=4}^n\epsilon_k$ for $n\geq 4.$
691:
692: In fact, if $M_1=1, M_2=2, M_3=3$, we have
693: $ \Theta^1(M)=(0,1,0,-1,(-1-R_n)_{n\geq 4}))$\\
694: \noindent If $M_1=1, M_2=0, M_3=-1$, then
695: $ \Theta^1(M)=(0,1,2,3,(3-R_n)_{n\geq 4}))$\\
696: \noindent If $M_1=-1, M_2=0, M_3=1$, then
697: $ \Theta^1(M)=(0,-1,0,1,(1-R_n)_{n\geq 4}))$\\
698: \noindent If $M_1=-1, M_2=-2, M_3=-3$, then
699: $ \Theta^1(M)=(0,-1,-2,-3,\Theta^1(-3-(R_n)_{n\geq 4}))$
700:
701: Similar presentation is valid for $M$:\\
702: if $M_1=1, M_2=2, M_3=3$, then
703: $ M =(0,1,2,3,(3+R_n)_{n\geq 4}))$,\\
704: if $M_1=1, M_2=0, M_3=-1$, then
705: $ M =(0,1,0,-1,(-1+R_n)_{n\geq 4}))$\\
706: if $M_1=-1, M_2=0, M_3=1$, then
707: $ M =(0,-1,0,1,(1+R_n)_{n\geq 4}))$\\
708: if $M_1=-1, M_2=-2, M_3=-3$, then
709: $ M =(0,-1,-2,-3,\Theta^1(-3+(R_n)_{n\geq 4}))$\\
710: To see that the laws of $\Theta^1(M)$ and $M$ are equal it is
711: convenient to pass to increments of corresponding processes.
712:
713: If we take a pass with $M_1=1, M_2=2, M_3=3$, then $\Theta^2(M)$ of
714: such trajectory has a probability zero which is not the case for the
715: corresponding trajectory of $M$. So, $\Theta²(M)\nel M$.
716: For $a\geq 3$ we can write that
717: $$\Theta^3(M) = (M_1, M_2, M_3, \Theta^3((M_k)_{k\geq 4}))$$
718: and we conclude from symmetry of Bernoulli random walk.\vskip 0.5cm
719:
720: \begin{center}
721: \includegraphics{Ocone1.eps}
722: \end{center}
723:
724: \vspace*{.5in}
725:
726: \noindent {\bf Counterexample 2}:
727: Let $(\varepsilon_k)_{k\ge0}$ be a sequence of independent $\{-1,+1\}$-valued symmetric Bernoulli
728: random variables. Set $k_n=\left\lfloor\frac{\ln (n+1)}{\ln 2}\right\rfloor-1$, where $\lfloor x\rfloor$
729: is the lower integer part of $x$ and let us consider the
730: following skip free process:
731: \[M_0=0\;\;\mbox{and for $n\ge1$,}\;\;M_n=\sum_{k=0}^{k_n} 2^k\varepsilon_k+(n-2^{k_n})\varepsilon_n\,.\]
732: Actually, $M$ is constructed as follows: $M_0=0$, $M_1=\varepsilon_0$ and for all $k\ge1$ and
733: $n\in[2^{k},2^{k+1}-1]$, the increments $M_n-M_{n-1}$
734: have the sign of $\varepsilon_k$. In particular, the increments of $(M_n)$ are $-1$ or $1$ and since,
735: from the discussion at the beginning of section \ref{skipfree}, the only skip
736: free local martingale with such increments is the Bernoulli random walk, it is clear that $M$ is not
737: an Ocone local martingale.\\
738:
739: \begin{center}
740: \includegraphics{Ocone2.eps}
741: \end{center}
742:
743: \vspace*{.2in}
744:
745: \noindent The equality $\Theta^0(M)\el M$ only means that $M$
746: is a symmetric process, which is straightforward from its construction. Now let us check that $T_1<\infty$,
747: a.s.~and $\Theta^1(M)\el M$. Almost
748: surely on the set $\{M_1=-1\}$, there exists $k\ge0$ such that
749: $\varepsilon_i=-1$ for all $i\le k$ and $\varepsilon_{k+1}=1$. The later assertion is equivalent to say that
750: for all integer $n\in(0,2^{k+1}-1]$, $M_n-M_{n-1}=-1$ and for all
751: $n\in[2^{k+1},2^{k+2}-1]$, $M_n-M_{n-1}=1$. It is then easy to check that
752: \[M_{2^{k+2}-1}=1\,.\]
753: So we have proved that $\{M_1=-1\}\subseteq\{T_{1}<\infty\}$, but since we also have
754: $\{M_1=1\}\subseteq\{T_{1}<\infty\}$, it follows that $\p(T_{1}<\infty)=1$.
755:
756: Then we see from the construction of $(M_n)$
757: that almost surely, $T_1$ belongs to the set $\{2^j-1:j\ge1\}$ and that for $j\ge1$, conditionally to
758: $T_1=2^j-1$, $(M_n,\,n\le T_1)$ and $(M_{T_1+n},\,n\ge0)$ are independent. Moreover,
759: \[(M_{T_1+n},\,n\ge0)\el (2-M_{T_1+n},\,n\ge0)\,,\]
760: so this proves that $\Theta^1(M)\el M$.\\
761:
762: Finally note that $0$ and $1$ are the only nonnegative levels at which the reflection principle
763: holds for the process $M$, i.e. $\Theta^a(M)\el M$ implies $a=0$ or $1$. Indeed, at least it is clear
764: from the
765: construction of $M$ that the only times and levels at which the sign of its increments can change belong
766: to the set $\{2^j-1,\,j\ge0\}$, i.e. if $a\ge0$ is such that $\Theta^a(M)\el M$, then necessarily
767: $a\in\{2^j-1,\,j\ge0\}$ and $T_a\in\{2^j-1,\,j\ge0\}$. But suppose that for $i\ge2$ we have $T_1=2^{i}-1$
768: and recall that all the increments $M_{T_1+k+1}-M_{T_1+k}$ for all $k=0,1,\dots 2^i-1$ have the same sign. If
769: these increments are 1, then the process $M$ reaches the level $2^i-1$ at time $T_1+2^i-2=2^{i+1}-3$ which
770: does not belong to the set $\{2^j-1,\,j\ge0\}$. So the sign of the increments of $M$ cannot change at this
771: time and the level $2^i-1$ cannot satisfy the identity in law $\Theta^{2^i-1}(M)\el M$.
772:
773:
774: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
775: \subsection{Continuous time lattice processes}\label{lattice}
776: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
777: As a preliminary result for the proof of Theorem \ref{main1}, we state an analogue of Theorem
778: \ref{prop2} for continuous time lattice processes.
779: We say that $M=(M_t)_{t\geq 0}$ is a {\it continuous time lattice process} if $M_0=0$ and if it is a
780: pure jump c\`adl\`ag process whose jumps $\Delta M_t= M_t-M_{t-}$
781: verify : $|\Delta M_t|=\eta$, for some fixed real $\eta>0$. If we denote by
782: $(\tau_k)_{k\geq 1}$ the jump times of $M$, i.e. with $\tau_0=0$, for $k\ge1$,
783: \[\tau_k=\inf\{t>\tau_{k-1}:|M_t-M_{\tau_{k-1}}|=\eta\}\,,\]
784: with $\inf\{\emptyset\}=\tau_{k-1}$, then for all $t\ge0$ and $\p$-a.s.
785: $$M_t=\sum_{k=1}^{\infty}\Delta M_{\tau_k}\ind _{\{\tau_k\leq t\}}\,.$$
786: The quadratic variation of $M$ is given by:
787: $$[M]_t=\sum_{k=1}^{\infty}(\Delta M_{\tau_k})^2\ind _{\{\tau_k\leq
788: t\}}=\eta^2\sum_{k=1}^{\infty}\ind _{\{\tau_k\leq t\}}.$$
789: Note that $\tau_k$ admits the equivalent definition $\tau_k=\inf\{t\ge0:[M]_t=k\eta^2\}$.
790: We define the time changed discrete process $S^M$ by $S^M=(M_{\tau_k})_{k\ge0}$ which has values
791: in the lattice $\eta\mathbb{Z}$. In particular, we have:
792: \begin{equation}\label{idlattice}
793: M_t=S^M_{\eta^{-2}[M]_t}\,,\quad t\ge0\,.
794: \end{equation}
795: We say that $M$ is a {\it continuous time lattice Ocone local martingale} if it can be written as
796: $M_t=S_{A_t}$, where $S$ is a symmetric Bernoulli random walk with values in the lattice $\eta\mathbb{Z}$
797: and $A$ is an increasing continuous time lattice process with values in $\mathbb{N}$ which is independent
798: of $S$. In the case where $M$ is divergent, $S$ coincide with $S^ M$ given in formula (\ref{idlattice}).
799: When $M$ is not divergent, $S$ is different from $S^ M$, namely if
800: $T= \inf \{k\geq 0: [S^ M]_k= [S^ M]_{\infty}\}$ then $S$ can be taken as:
801: $$S_k= S^M_k \ind_{\{ k\leq T\}} + (S^M_T + \tilde{S}_{T-k})\ind_{\{ k>T\}}\,,$$
802: where $\tilde{S}$ is a symmetric Bernoulli random walk which is independent from $S^M$.
803: In this case $S$ is independent from $[M]$. Therefore, when considering a continuous time lattice Ocone
804: local martingale $M$, in identity (\ref{idlattice}) we can and will suppose that $S^M$ is a symmetric Bernoulli
805: random walk with values in the lattice $\eta\mathbb{Z}$ and which is independent of $[M]$.
806: \vspace{0.2cm}
807:
808: Recall the definitions (\ref{ta}) and (\ref{transf}) of the hitting
809: time $T_{a}$ and transformations $\Theta^{a}$, respectively.
810:
811: \begin{proposition}\label{c} Let $M$ be any continuous time lattice process
812: such that for all $k=0,1,2$, $$\Theta^{k\eta}(M)\el M\,,$$
813: then $M$ is a continuous time lattice Ocone local martingale.
814: If in addition $T_{\eta}< \infty$ a.s.,
815: then $S^M$ is a symmetric random walk on the lattice
816: $\eta\mathbb{Z}$ which is independent of $[M]$. Moreover, $M$ is a divergent local martingale
817: with respect to its own filtration.
818: \end{proposition}
819: \noindent {\it Proof}. Set $N=\eta^{-1}M$. We remark that for $k=1,2,3$,
820: $$\Theta^{k}(N)\el N.$$ Then following the proof of
821: Theorem \ref{prop2} along the lines for the continuous time process $N$, we obtain that $S^N$
822: conditionally to $[N]$ is Bernoulli random walk. Hence $S^{N}$ is a Bernoulli random walk which
823: is independent of $[N]$. Since $S^N=\eta^{-1}S^M$ and
824: $\eta^{-2}[N]=[M]$, we obtain that $S^M$ is a symmetric Bernoulli
825: random walk on the lattice $\eta\mathbb{Z}$
826: which is independent of $[M]$. It means that it is local
827: martingale with respect to its own filtration. Finally, when $T_{\eta}<\infty$ a.s., $M$
828: is a divergent local martingale since $N$ is so.\QED
829:
830:
831: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
832: \section{Proof of theorem 1}\label{proof}
833: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
834:
835: Let $(\Omega ,\mathcal F , \mathbb F, \p)$ be the canonical space of continuous functions
836: with filtration $\mathbb F$ satisfying usual conditions.
837: Let $M$ be a continuous stochastic process which is defined on this space and
838: satisfying the assumptions of Theorem~\ref{main1}. Without loss of generality we
839: suppose that the sequence $(a_n)$ is decreasing.
840: \vskip 0.5cm
841:
842: \noindent {\it Proof of Theorem $\ref{main1}$}. First of all we note
843: that since the map $x\rightarrow \Theta^x(\omega)$ is continuous on
844: $C(\R^+, \R)$, the hypothesis of this
845: theorem imply that $\Theta^0(M)\stackrel{\mathcal L}{=}M$, i.e. $M$ is
846: symmetric process.
847:
848: Now, fix a positive integer $n$. We define the continuous lattice valued process
849: $M^n$ by using discretisation with respect to the space variable. In this aim, we
850: introduce the sequence
851: of stopping times $(\tau_k^n)_{k\geq 0}$ i.e. $\tau_0^n=0$ and for all $k\ge1$
852: $$\tau ^n_k= \inf\{t>\tau ^n_{k-1} \,: \, |M_t-M_{\tau ^n_{k-1}}| =a_n\}\,,$$
853: with $\inf\{\emptyset \}=\tau^n_{k-1}$. Then $M^n=(M^n_t)_{t\geq0}$ is defined by:
854: $$M^n_t= \sum^{\infty}_{k=0} M_{\tau ^n_{k}}
855: \ind _{\{\tau^n_{k}\leq t < \tau ^n_{k+1}\}}\,.$$ We can easily check that $M^n$
856: is a continuous time lattice process verifying the assumptions of Proposition~\ref{c}.
857: Therefore according to this proposition, $[M^n]$ is a continuous time lattice Ocone
858: local martingale.
859:
860: From the construction of $M^n$ we have the almost sure inequality
861: \begin{equation}\label{bound}
862: \sup _{t\geq 0}|M_t -M^n_t|\leq a_n\,.
863: \end{equation}
864: Hence the sequence $(M^n)$ converges a.s.~uniformly on $[0,\infty)$ toward $M$.
865: The condition $$\sup_{n\geq 1}\sup_{t\geq 0}|\Delta M^n_t|\leq a_1$$
866: and (\ref{bound}) imply (cf. \cite{JS},Corollary IX.1.19, Corollary
867: VI.6.6) that $M$ is a local martingale and that
868: \begin{equation}\label{conv}
869: (M^n, [M^n])\stackrel{\mathcal L}{\rightarrow}(M,\langle M\rangle)\,.
870: \end{equation}
871: Since the properties (i) and (iii) given in introduction are equivalent, it is sufficient to verify that
872: for every deterministic function $h$ of the form $\sum_{j=1}^k\lambda_j\ind_{]t_{j-1}, t_{j}]}$
873: with $t_0=0 <t_1<\cdots t_k$ we have:
874: \begin{equation}\label{iii}
875: \e\left[\exp\left(i\int_0^\infty h(s)\,dM_s\right)\right]=
876: \e\left[\exp\left(-\frac12\int_0^\infty h^2(s)\,d\langle
877: M\rangle_s\right)\right]\,.
878: \end{equation}
879: From (\ref{conv}) we see that
880: $$\lim _{n\rightarrow \infty}\e\left[\exp\left(i\int_0^\infty h(s)\,dM_s^n\right)\right] =
881: \e\left[\exp\left(i\int_0^\infty h(s)\,dM_s\right)\right]\,.$$
882: Then in order to obtain (\ref{iii}), we will show by straightforward calculations that
883: \begin{equation}\label{calc}
884: \lim _{n\rightarrow \infty}\e\left[\exp\left(i\int_0^\infty h(s)\,dM_s^n\right)\right]=
885: \e\left[\exp\left(-\frac12\int_0^\infty h^2(s)\,d\langle
886: M\rangle_s\right)\right]\,.
887: \end{equation}
888: To prove (\ref{calc}) we first write
889: \[\e\left[\exp\left(i\int_0^\infty h(s)\,dM_s^n\right)\right]=
890: \int\e\left[\exp\left(i\int_0^\infty h(s)\,dM_s^n\right)\,|\,[M^n]=\omega\right]
891: \,dP_{[M^n]}(\omega)\,,\]
892: where $P_{[M^n]}$ is the law of $[M^n]$.
893: Then from Proposition 1 we have that
894: $$M^n \stackrel{\mathcal L}{=}a_nS_{a_n^{-2}[M^n]}$$
895: where $S$ is symmetric Bernoulli random walk independent from $[M^n]$.
896: Moreover,
897: $$\int_0^\infty h(s)\,dM_s^n = \sum _{j=1}^{k} \lambda _j \Delta M_{t_j}^n
898: \stackrel{\mathcal L}{=}a_n\sum _{j=1}^{k} \lambda _j \Delta S_{r_j}$$
899: where $\Delta M_{t_j}^n = M_{t_{j}}^n- M_{t_{j-1}}^n$, $\Delta S_{r_j}= S_{r_{j}} - S_{r_{j-1}}$ and
900: $r_j= a_n^{-2}[M^n]_{t_j}$, $1\le j\le k$.\\
901:
902: Since $S$ and $[M^n]$ are independent and $\e\left[\exp (i a \Delta S_k)\right] = \cos (a)$ for all $a\in \R$, we have:
903: \begin{equation}\label{podivon}
904: \e\left[\exp\left(i\int_0^\infty h(s)\,dM_s^n\right)\,|\, [M^n]=\omega\right]=
905: \prod_{j=1}^k [\cos(\lambda _j a_n)]^{(u_j^n -u_{j-1}^n)}\,,
906: \end{equation}
907: where $u^n_j=\lfloor a_n^{-2}\omega_{t_j}\rfloor$, $j=0,1,\dots,k$ and $\lfloor x\rfloor$ is the lower integer part
908: of $x$. Moreover, it is not difficult to see that
909: \begin{equation}\label{hvatit}
910: \lim _{n\rightarrow \infty}\prod_{j=1}^k [\cos(\lambda _j a_n)]^{(u_j^n -u_{j-1}^n)} =
911: \exp\left(-\frac{1}{2}\sum_{j=1}^k \lambda_j^2(\omega_{t_j} - \omega_{t_{j-1}})\right)
912: \end{equation}
913: uniformly on compact sets of $\R_+^k$.
914: Then, the expression (\ref{podivon}) and the convergence relations (\ref{conv}), (\ref{hvatit}) imply (\ref{calc}). \QED
915:
916:
917: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
918: \section{Proof of Theorem \ref{main2}}\label{conj}
919: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
920: In what follows we assume, without loss of generality, that the process $M$ is divergent.
921: We begin with the following classical result of ergodic theory, a proof of which may be
922: found in \cite{dey}, Lemma 1.
923:
924: Let $(\Omega ,\mathcal F , \mathbb F, \p)$ be canonical space of
925: continuous functions endowed by natural right-continuous filtration $\mathbb F =
926: (\mathcal F _t)_{t\geq 0}$ completed by negligible sets of $\mathcal
927: F= \bigvee_{t\geq 0} \mathcal F_t$.
928:
929: \begin{lemma}\label{lemdey}
930: Let $\Theta$ be a measurable transformation
931: of $\Omega$ to $\Omega$ which preserves $\p$. A random variable $X\in L^2(\Omega,{\cal F},\p)$ is
932: a.s.~invariant by $\Theta$ if and only if
933: \[\e (Z\cdot(Y\circ \Theta))=\e (Z\cdot Y)\,,\]
934: for all $Y\in L^2(\Omega,{\cal F},\p)$.
935: \end{lemma}\vskip 0.5cm
936:
937: Let $\Theta_n$, $n\ge1$ be a family of
938: transformations defined on canonical space of continuous functions
939: $(\Omega, \mathcal F, \mathbb F,\p)$.
940: Let ${\cal I}$ be the sub $\sigma$-algebra of the invariant events by all the transformations
941: $\Theta_n$, $n\ge1$, i.e.
942: \[{\cal I}=\{F\in{\cal F}:\ind_{F}\circ \Theta_n \stackrel{\mbox{a.s.}}{=}\ind_F,\;\mbox{for all $n\ge1$}\}.\]
943:
944: The following lemma extends Theorem 1 in \cite{dey}.
945: \begin{lemma}\label{lemcondind}
946: Let $M$ be a continuous divergent local martingale defined on the filtered probability space
947: $(\Omega,\mathbb F, {\cal F},\p)$. Assume that the transformations $\Theta_n$ preserve the Wiener measure,
948: i.e.~if $B$ is the standard Brownian motion then for all $n\ge1$,
949: $B\circ \Theta_n\el B$.
950: The following assertions are equivalent:
951: \begin{itemize}
952: \item[$(j)$] For all $n\ge1$, $(B^M,\langle M\rangle)\el(\Theta_n(B^M),\langle M\rangle)$ have the same law.
953: \item[$(jj)$] $B^M$ and $\langle M\rangle$ are conditionally independent given the $\sigma$-field
954: $\mathcal I^M = (B^M)^{-1}({\cal I})$.
955: \end{itemize}
956: \end{lemma}
957: \noindent {\it Proof}. The proof almost follows from this of Theorem 1 in \cite{dey} along the lines.
958: We first prove that $(j)$ implies $(jj)$.
959:
960: Let $h,g$ two measurable functions $\Omega \rightarrow
961: \Omega$. Then (j) implies:
962: \begin{equation}\label{fu1}
963: \e\left(h(\langle M\rangle) g(B^M)\right)= \e\left(h(\langle M\rangle) g(B^M\circ
964: \Theta_n)\right) =\e\left(h(\langle M\rangle)( g(B^M)\circ \Theta_n)\right)
965: \end{equation}
966: We take conditional expectation with respect to $B^M$. For this we
967: denote by $f$ the following function:
968: $$\e \left(h(\langle M\rangle) | B^M\right)\stackrel{\mbox{a.s.}}{=}f(B^M)$$
969: Then (\ref{fu1}) implies that
970: \begin{equation}\label{fu2}
971: \e\left(f(B^M) g(B^M)\right)= \e\left(f(B^M)( g(B^M)\circ \Theta_n)\right)
972: \end{equation}
973: Then according to Lemma \ref{lemdey} $f(B^M)$ is $\Theta_n$-invariant
974: variable, i.e. it is measurable with respect to $\sigma$-algebra of
975: $\Theta _n$-invariant sets $\mathcal I_n$. Since it holds for all $n\geq
976: 1$, $f(B^M)$ is measurable with respect to $\mathcal I=
977: \cap_{n=1}^{\infty}\mathcal I_n$.
978: Moreover,\\\\
979: $\e\left(h(\langle M\rangle) g(B^M)| \mathcal I^M\right)=
980: \e\left(f(B^M) g(B^M) | \mathcal I\right)=$
981: $$\e \left(f(B^M) | \mathcal I\right)\e \left( g(B^M) | \mathcal I\right)=
982: \e (h(\langle M\rangle) | \mathcal I^M)\e(g(B^M) | \mathcal I^M)$$
983: and $(jj)$ is proved.
984:
985: Now suppose that $(jj)$ is valid. Then
986: $$\e (h(\langle M\rangle) g(B^M))=
987: \e \left(\e \left(h(\langle M\rangle) | \mathcal
988: I^M\right)\e \left(g(B^M) | \mathcal I^M \right)\right) $$
989:
990: Moreover, since $B^M\circ \Theta_n$ and $\langle M \rangle $ are also conditionally independent for all
991: $n\ge1$, we have
992: $$\e \left(h(\langle M\rangle) g(B^M\circ \Theta_n)| \mathcal I^M) \right)
993: =\e \left(h(\langle M\rangle) | \mathcal
994: I^M\right)\e \left(g(B^M)\circ \Theta_n | \mathcal I^M\right) $$
995: Since every $ \mathcal I^M$-measurable random variable has the form $u(B^M)$,
996: where $u$ is ${\cal I}$-measurable,
997: $$\e\left(g(B^M)\circ \Theta_n)\,|\, \mathcal I^M\right) = \e\left(g(B^M)\,|\,
998: \mathcal I^M\right)$$
999: and we obtain
1000: $$\e \left(h(\langle M\rangle) g(B^M)\right)= \e\left(h(\langle M\rangle) g(B^M\circ
1001: \Theta_n)\right)$$
1002: which is $(j)$.\QED
1003:
1004: \noindent\it Proof of Theorem \ref{main2}.\rm
1005:
1006: If $(ii)$ holds then from Lemma \ref{lemcondind}, $B^M$ and $\langle M\rangle$ are
1007: independent, so $(i)$ holds. Let us prove that $(i)$ implies $(ii)$. Suppose that $(ii)$ fails.
1008: We show that $(i)$ fails, too.
1009: Namely we show that one can construct a continuous martingale $M = B_{A}$, where $B$ is standard
1010: Brownian motion and $A$ is non-decreasing
1011: continuous adapted process, such that $M$ verify reflection properties of $(i)$
1012: but it is not Ocone martingale.
1013:
1014: Let $X$ be a non trivial
1015: $B^{-1}({\cal I}^{a})$-measurable bounded random variable. Call $({\cal F}_t^B)$
1016: the natural filtration generated by $B$. Let $N_t=\e(X\,|\,{\cal F}_t^B) $ for all $t\geq 0$ and
1017: $N=(N_t)_{t\geq 0}$. We remark that $N$ is a $({\cal
1018: F}^B_t)$-martingale invariant by all transformations
1019: $(\Theta^{a_n})$:
1020: $$N \stackrel{\mathcal L}{=} N \circ \Theta^{a_n}.$$
1021:
1022: Now, we can construct a finite non-constant stopping time
1023: $T$ which is invariant by all the transformations $\Theta^{a_n}$ by
1024: setting $\,T=\inf \{t\geq t_0 \,|\, N_t\in K\}$, where $t_0$ is large enough and
1025: $K$ is a suitable Borel set. For instance we can choose $K$ such that
1026: $\p(X\in K)\ge2/3$. Since $N_t\rightarrow X$ a.s. as $t\rightarrow
1027: \infty$ we can find $t_0$ such that for $t\geq t_0$, $\p(N_t\in K)\ge1/2$.
1028:
1029: Finally, for $\alpha>0$, let us define
1030: the following increasing process
1031: \[A_t=\int_0^t\ind_{[0,T]}(s)+\alpha\ind_{]T,\infty[}(s)\,ds\,.\]
1032: This process is not deterministic whenever $\alpha\neq1$ and since it is invariant by all the transformations
1033: $\Theta^{a_n}$, one has $(B,A)\el(\Theta^{a_n}(B),A)$ for all $n\ge1$. The inverse of $A$ is given by
1034: \[A_t^{-1}=\int_0^t\ind_{[0,T]}(s)+\alpha^{-1}\ind_{]T,\infty[}(s)\,ds\,,\]
1035: so it is adapted and each $A_t$ is a $({\cal F}^B_t)$-stopping time.
1036:
1037: Therefore $M=(M_t)_{t\geq 0}$ with $M_t=B_{A_t}$ is a continuous
1038: divergent $({\cal F}^B_{A_t})$-martingale satisfying $\Theta^{a_n}(M)\el M$, for all $n\ge1$. Moreover, $B^M=B$
1039: and $\langle M\rangle=A$ are not independent by construction. Hence, $M$ can not be Ocone martingale with respect to the filtration
1040: $({\cal F}^B_{A_t})_{t\geq 0}$ and it provides
1041: a counterexample to the assertion $(i)$. So, we have proved that $(i)$ implies $(ii)$.\QED
1042:
1043: \vspace*{.2in}
1044:
1045: \begin{thebibliography}{99}
1046:
1047: \bibitem{an} \sc D. Andr\'e: \rm Solution directe du probl\`eme r\'esolu par M.~Bertrand.
1048: {\it C.R. Acad. Sci. Paris}, {\bf 105}, 436-437, (1887).
1049:
1050: \bibitem{oc} \sc D.L.~Ocone: \rm A symmetry characterization of conditionally
1051: independent increment martingales. \it Barcelona
1052: Seminar on Stochastic Analysis \rm 147--167, Progr. Probab., {\bf 32},
1053: Birkh\"auser, Basel, 1993.
1054:
1055: \bibitem{dey} \sc L.E.~Dubins, M.~\'Emery and M.~Yor: \rm
1056: On the L\'evy transformation of Brownian motions and continuous martingales.
1057: \it S\'eminaire de Probabilit\'es, \rm XXVII, 122--132, Lecture
1058: Notes in Math., {\bf 1557}, Springer, Berlin, 1993.
1059:
1060: \bibitem{ds} \sc L.E.~Dubins and M.~Smorodinsky: \rm The modified, discrete,
1061: L\'evy-transformation is Bernoulli. {\it S\'eminaire de
1062: Probabilit\'es}, XXVI, 157--161, Lecture Notes in Math., {\bf 1526},
1063: Springer, Berlin, 1992.
1064:
1065: \bibitem{JS} \sc J.~Jacod and A.~Shiryaev: \it Limit Theorems for
1066: Stochastic Processes, \rm Springer-Verlag Berlin, Heidelberg New York, 1987.
1067:
1068: \bibitem{ma1} \sc M.~Malric: \rm Transformation de L\'evy et z\'eros du
1069: mouvement brownien. {\it Probab. Theory Related Fields}, {\bf 101},
1070: no. 2, 227--236, (1995).
1071:
1072: \bibitem{ma2} \sc M.~Malric: \rm Densit\'e des z\'eros des transform\'es de L\'evy
1073: it\'er\'es d'un mouvement brownien. {\it C. R. Math. Acad. Sci.
1074: Paris}, {\bf 336}, no. 6, 499--504, (2003).
1075:
1076: \bibitem{vy} \sc L.~Vostrikova and M.~Yor: \rm
1077: Some invariance properties (of the laws) of Ocone's martingales. \it
1078: S\'eminaire de Probabilit\'es, \rm XXXIV, 417--431, Lecture Notes in
1079: Math., {\bf 1729}, Springer, Berlin, 2000.
1080:
1081: \end{thebibliography}
1082:
1083: \end{document}
1084:
1085: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1086: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1087: %%%%%%%%%%%%%%%%
1088: